Model overconfidence and poor calibration are common in machine learning and difficult to account for when applying standard empirical risk minimization. In this work, we propose a novel method to alleviate these problems that we call odd-$k$-out learning (OKO), which minimizes the cross-entropy error for sets rather than for single examples. This naturally allows the model to capture correlations across data examples and achieves both better accuracy and calibration, especially in limited training data and class-imbalanced regimes. Perhaps surprisingly, OKO often yields better calibration even when training with hard labels and dropping any additional calibration parameter tuning, such as temperature scaling. We provide theoretical justification, establishing that OKO naturally yields better calibration, and provide extensive experimental analyses that corroborate our theoretical findings. We emphasize that OKO is a general framework that can be easily adapted to many settings and the trained model can be applied to single examples at inference time, without introducing significant run-time overhead or architecture changes.
In recent years, great progress has been made in the Lift-Splat-Shot-based (LSS-based) 3D object detection method, which converts features of 2D camera view and 3D lidar view to Bird's-Eye-View (BEV) for feature fusion. However, inaccurate depth estimation (e.g. the 'depth jump' problem) is an obstacle to develop LSS-based methods. To alleviate the 'depth jump' problem, we proposed Edge-Aware Bird's-Eye-View (EA-BEV) projector. By coupling proposed edge-aware depth fusion module and depth estimate module, the proposed EA-BEV projector solves the problem and enforces refined supervision on depth. Besides, we propose sparse depth supervision and gradient edge depth supervision, for constraining learning on global depth and local marginal depth information. Our EA-BEV projector is a plug-and-play module for any LSS-based 3D object detection models, and effectively improves the baseline performance. We demonstrate the effectiveness on the nuScenes benchmark. On the nuScenes 3D object detection validation dataset, our proposed EA-BEV projector can boost several state-of-the-art LLS-based baselines on nuScenes 3D object detection benchmark and nuScenes BEV map segmentation benchmark with negligible increment of inference time.
We study the setting of optimizing with bandit feedback with additional prior knowledge provided to the learner in the form of an initial hint of the optimal action. We present a novel algorithm for stochastic linear bandits that uses this hint to improve its regret to $\tilde O(\sqrt{T})$ when the hint is accurate, while maintaining a minimax-optimal $\tilde O(d\sqrt{T})$ regret independent of the quality of the hint. Furthermore, we provide a Pareto frontier of tight tradeoffs between best-case and worst-case regret, with matching lower bounds. Perhaps surprisingly, our work shows that leveraging a hint shows provable gains without sacrificing worst-case performance, implying that our algorithm adapts to the quality of the hint for free. We also provide an extension of our algorithm to the case of $m$ initial hints, showing that we can achieve a $\tilde O(m^{2/3}\sqrt{T})$ regret.
With the rapid increase in the integration of renewable energy generation and the wide adoption of various electric appliances, power grids are now faced with more and more challenges. One prominent challenge is to implement efficient anomaly detection for different types of anomalous behaviors within power grids. These anomalous behaviors might be induced by unusual consumption patterns of the users, faulty grid infrastructures, outages, external cyberattacks, or energy fraud. Identifying such anomalies is of critical importance for the reliable and efficient operation of modern power grids. Various methods have been proposed for anomaly detection on power grid time-series data. This paper presents a short survey of the recent advances in anomaly detection for power grid time-series data. Specifically, we first outline current research challenges in the power grid anomaly detection domain and further review the major anomaly detection approaches. Finally, we conclude the survey by identifying the potential directions for future research.
The 3D building modelling is important in urban planning and related domains that draw upon the content of 3D models of urban scenes. Such 3D models can be used to visualize city images at multiple scales from individual buildings to entire cities prior to and after a change has occurred. This ability is of great importance in day-to-day work and special projects undertaken by planners, geo-designers, and architects. In this research, we implemented a novel approach to 3D building models for such matter, which included the integration of geographic information systems (GIS) and 3D Computer Graphics (3DCG) components that generate 3D house models from building footprints (polygons), and the automated generation of simple and complex roof geometries for rapid roof area damage reporting. These polygons (footprints) are usually orthogonal. A complicated orthogonal polygon can be partitioned into a set of rectangles. The proposed GIS and 3DCG integrated system partitions orthogonal building polygons into a set of rectangles and places rectangular roofs and box-shaped building bodies on these rectangles. Since technicians are drawing these polygons manually with digitizers, depending on aerial photos, not all building polygons are precisely orthogonal. But, when placing a set of boxes as building bodies for creating the buildings, there may be gaps or overlaps between these boxes if building polygons are not precisely orthogonal. In our proposal, after approximately orthogonal building polygons are partitioned and rectified into a set of mutually orthogonal rectangles, each rectangle knows which rectangle is adjacent to and which edge of the rectangle is adjacent to, which will avoid unwanted intersection of windows and doors when building bodies combined.
Deep learning recommendation systems must provide high quality, personalized content under strict tail-latency targets and high system loads. This paper presents RecPipe, a system to jointly optimize recommendation quality and inference performance. Central to RecPipe is decomposing recommendation models into multi-stage pipelines to maintain quality while reducing compute complexity and exposing distinct parallelism opportunities. RecPipe implements an inference scheduler to map multi-stage recommendation engines onto commodity, heterogeneous platforms (e.g., CPUs, GPUs).While the hardware-aware scheduling improves ranking efficiency, the commodity platforms suffer from many limitations requiring specialized hardware. Thus, we design RecPipeAccel (RPAccel), a custom accelerator that jointly optimizes quality, tail-latency, and system throughput. RPAc-cel is designed specifically to exploit the distinct design space opened via RecPipe. In particular, RPAccel processes queries in sub-batches to pipeline recommendation stages, implements dual static and dynamic embedding caches, a set of top-k filtering units, and a reconfigurable systolic array. Com-pared to prior-art and at iso-quality, we demonstrate that RPAccel improves latency and throughput by 3x and 6x.
Clinical question answering (QA) aims to automatically answer questions from medical professionals based on clinical texts. Studies show that neural QA models trained on one corpus may not generalize well to new clinical texts from a different institute or a different patient group, where large-scale QA pairs are not readily available for retraining. To address this challenge, we propose a simple yet effective framework, CliniQG4QA, which leverages question generation (QG) to synthesize QA pairs on new clinical contexts and boosts QA models without requiring manual annotations. In order to generate diverse types of questions that are essential for training QA models, we further introduce a seq2seq-based question phrase prediction (QPP) module that can be used together with most existing QG models to diversify their generation. Our comprehensive experiment results show that the QA corpus generated by our framework is helpful to improve QA models on the new contexts (up to 8% absolute gain in terms of Exact Match), and that the QPP module plays a crucial role in achieving the gain.
Knowledge distillation (KD) is an effective technique to transfer knowledge from one neural network (teacher) to another (student), thus improving the performance of the student. To make the student better mimic the behavior of the teacher, the existing work focuses on designing different criteria to align their logits or representations. Different from these efforts, we address knowledge distillation from a novel data perspective. We argue that transferring knowledge at sparse training data points cannot enable the student to well capture the local shape of the teacher function. To address this issue, we propose locally linear region knowledge distillation ($\rm L^2$RKD) which transfers the knowledge in local, linear regions from a teacher to a student. This is achieved by enforcing the student to mimic the outputs of the teacher function in local, linear regions. To the end, the student is able to better capture the local shape of the teacher function and thus achieves a better performance. Despite its simplicity, extensive experiments demonstrate that $\rm L^2$RKD is superior to the original KD in many aspects as it outperforms KD and the other state-of-the-art approaches by a large margin, shows robustness and superiority under few-shot settings, and is more compatible with the existing distillation approaches to further improve their performances significantly.
Knowledge distillation (KD) is an effective technique to transfer knowledge from one neural network (teacher) to another (student), thus improving the performance of the student. The existing work trains the student to mimic the outputs of the teacher on training data. We argue that transferring knowledge at sparse training data points cannot enable the student to well capture the local shape of the teacher's function. To address this issue, we propose locally linear region knowledge distillation ($\rm L^2$RKD) which transfers the knowledge in local, liner regions from a teacher to a student. $\rm L^2$RKD enforces the student to mimic the local shape of the teacher function in linear regions. Extensive experiments with various network architectures demonstrate that $\rm L^2$RKD outperforms the state-of-the-art approaches by a large margin and is more data-efficient. Moreover, $\rm L^2$RKD is compatible with the existing distillation methods and further improves their performances significantly.
It is well observed that in deep learning and computer vision literature, visual data are always represented in a manually designed coding scheme (eg., RGB images are represented as integers ranging from 0 to 255 for each channel) when they are input to an end-to-end deep neural network (DNN) for any learning task. We boldly question whether the manually designed inputs are good for DNN training for different tasks and study whether the input to a DNN can be optimally learned end-to-end together with learning the weights of the DNN. In this paper, we propose the paradigm of {\em deep collective learning} which aims to learn the weights of DNNs and the inputs to DNNs simultaneously for given tasks. We note that collective learning has been implicitly but widely used in natural language processing while it has almost never been studied in computer vision. Consequently, we propose the lookup vision networks (Lookup-VNets) as a solution to deep collective learning in computer vision. This is achieved by associating each color in each channel with a vector in lookup tables. As learning inputs in computer vision has almost never been studied in the existing literature, we explore several aspects of this question through varieties of experiments on image classification tasks. Experimental results on four benchmark datasets, i.e., CIFAR-10, CIFAR-100, Tiny ImageNet, and ImageNet (ILSVRC2012) have shown several surprising characteristics of Lookup-VNets and have demonstrated the advantages and promise of Lookup-VNets and deep collective learning.